There has been a long-standing notion in the materials science community that materials’ functional properties are strongly tied to the underlying defect substructure. Metals, for instance, consist of complex hierarchical networks of crystalline defects (i.e. vacancies, interstitials, dislocation loops, and grain boundaries) that have a strong bearing on the associated mechanical response (hardness, plasticity, fracture toughness, creep properties). Nevertheless, the nature of such inherent microstructure-property correlations under extreme conditions remains elusive to this date. In response to high deformation rates, elevated temperatures, and/or high-dose irradiation effects, crystalline flaws often interplay and evolve in highly nonlinear and stochastic ways, making the property prediction based on structural metrics a formidable task. Empirical frameworks conventionally describe these correlations by a fairly small set of "descriptors" largely ignoring inherent scale hierarchies and intricate topology of defect networks at micro/nano-structural levels. Multi-scale simulations have fairly limited applicability/predictability due to modeling gaps in transferring physics-based information across length/time-scales. Experimental investigations can only explore a small portion of the immense combinatorial space spanned by varying environments and different elemental compositions.

The above limitations demand applications of machine learning (ml) that can help establish robust relationships between defects’ heterogeneous microstructure and materials’ response within a "microstructural informatics" framework. The latest developments include deep-learned data mining for feature extraction, neural net-based interatomic potentials for complex defects, and graph network representations of heterogeneous microstructures. Obvious questions and challenges have yet to be fully addressed: 1) accurate identification and classification of topological defects through robust ml-based metrics that fully account for associated spatio-temporal variations under extreme conditions 2) construction of efficient ml force fields for strongly interacting defects to model their collective behavior with ab-initio accuracy but beyond atomistic scales 3) applications of ml to bridge existing gaps across scales in physics-based simulations to accelerate the design process of heterogeneous materials and microstructural tailoring 4) leverage the notion of "tractability" and "interpretability" given the multi-combinatorial descriptors’ phase space via effective reduced-complexity models and feature engineering leading to the extraction of fundamental physics and underlying mechanisms. To address these challenges, the proposed symposium will aim to conduct a thorough survey of the current state-of-the-art in data mining and pattern detection, feature extraction and analysis, and interpretation of ml predictions relevant to defects’ characterization and associated physics under harsh environments. We invite relevant contributions from academia and industry employing advanced computational/experimental techniques powered by ml to explore microstructure-property correlations in a broad range of contexts including chemically complex alloys and composites, amorphous particulate systems, metallic glasses, two-dimensional heterostructures and irradiated materials.

2D materials alloy diffusion fracture glass strain relationship superplasticity